Flow: create and run a local runbook

Save and verify the API requests, SSH Tasks and connections, and Database connections you intend to use. Start with a test environment and a small set of checks whose expected outcomes you know.

1. 1 Open Flow in the selected workspace and create a new definition.
2. 2 Name the runbook and declare its inputs: string, number, boolean, or JSON. Mark sensitive inputs secret; do not save secret defaults.
3. 3 Add API, SSH Task, or Database steps referencing saved resources. Choose the SSH connection explicitly and review the Database catalog and schema.
4. 4 Use Condition to branch on a comparison. For Wait Until, choose a GET or HEAD request, or a read-only query on a connection configured as read-only. Set the interval, timeout, and success or failure condition.
5. 5 Save the definition and resolve validation errors before running it.

Inputs and earlier step results can feed later steps. Text substitutions do not automatically escape SQL or shell commands; use trusted values and review the resulting operation.

1. 1 Open the run dialog and provide the required typed inputs.
2. 2 The desktop run dialog defaults to the active workspace environment. Review or change it for this run; choosing no environment uses workspace variables only.
3. 3 Review the Flow, workspace, environment, inputs, and possible effects in the confirmation, then start the run.
4. 4 Inspect step results and progress. Open History to review outcomes and Wait Until attempts, or select a previous run.
5. 5 Cancel if needed, then inspect any effects already completed. Cancellation does not roll them back.

Steps run serially. The first error stops later steps; ordinary action steps have no implicit retries. Wait Until performs only the checks and retry policy you define within its timeout.

Configure Codex or Cursor with the local Unfour MCP server. Flow tools share the desktop definitions, resources, and history: unfour.flow.list, get, save, delete, run, cancel_run, list_runs, and get_run. MCP history listing supports a limit and cursor.

• ✓ Pro Cloud Sync includes Flow definitions: steps, input declarations, and resource references. Run history, actual execution inputs/outputs, credentials, and device-local configuration stay local. Workspace import/export also includes Flow definitions, but excludes run history.
• ✓ Flow is a free core module, not a general workflow or autonomous troubleshooting platform.
• ✓ Saved API request scripts and multipart bodies are unsupported in Flow V1. Wait Until API probes must use GET/HEAD. Database probes require both a read-only connection and a read-only query.
• ✓ V1 has no scheduler, webhook triggers, parallel steps, sub-flows, or LLM/Agent nodes.
• ✓ Review any outputs sent to an MCP client: arbitrary logs and SQL results can contain confidential text.